Vmware Horizon View Large Scale Reference Architecture

VMware® Horizon View™ Large-Scale Reference ArchitectureR E F E R E N C E A R C H I T E C T U R E

VMware Horizon View Large-Scale Reference Architecture

R E F E R E N C E A R C H I T E C T U R E / 2

Table of Contents

Executive Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 VCE Vblock Specialized System for Extreme Applications . . . . . . . . . . . . . . . . . . . . . . 4 VMware Horizon View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Storage Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 EMC XtremIO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 EMC Isilon NAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Compute and Networking Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Workload Generation and Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Login VSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 VSImax and Horizon View Session Concurrency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Login VSI Test 1: 7,000-Seat Test with 100 Percent Session Concurrency . . . . . . 6 Login VSI Test 2: 7,000-Seat Test with 80 Percent Session Concurrency . . . . . . . 8 Timing Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Linked-Clone Desktop Pool Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Linked-Clone Desktop Pool Recompose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Storage Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11System Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 vSphere Cluster Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Networking Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 IP Network Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Storage Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 X-Brick Storage Configurations in vSphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 vSphere Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Infrastructure and Management Servers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Horizon View Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Basic Horizon View Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Test Pool Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Desktop Pool Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Cluster to Datastore Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Test Image Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 EMC XtremIO Storage Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 vStorage API for Array Integration (VAAI) Settings . . . . . . . . . . . . . . . . . . . . . . . . 24 HBA Queue Depth Adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Native vSphere Storage Multipathing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25About the Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27Appendix A (Login VSI Detailed Scoring Methodology) . . . . . . . . . . . . . . . . . . . . . . . . 28Appendix B (Test Methodology) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Deployment and Recompose Timing Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Login VSI Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Appendix C (Bill of Materials for Test Configuration) . . . . . . . . . . . . . . . . . . . . . . . . . . . 30



Executive SummaryThis reference architecture is based on real-world test scenarios, user workloads, and infrastructure system configurations. It uses the VCE Vblock Specialized System for Extreme Applications, composed of Cisco UCS server blades and EMC XtremIO flash storage, to support a 7,000-user VMware® Horizon View™ 5.2 deployment.

Extensive user-experience and operations testing, including Login VSI desktop performance benchmark testing on over 80,000 desktops, revealed world-class operational performance. A simple design architecture and efficient use of storage preserved ease of use at an attractive price point. The results are summarized here and further described later in this paper.

1,000 Linked ClonesDeployed

1 Hour15 Minutes

1 Hour45 Minutes

15 Minutes

6,432

0.9ms

3.5TB

1,000 Linked ClonesRecomposed

1,000 Linked ClonesStart-Up

Login VSI Max- 7,000 Active Sessions- Represents 100% Session Concurrency

Front-End DatastoreLatency

Total StorageFootprint

Figure 1: Highlights

Outstanding Desktop Performance

•Desktopsexhibitedperformancecharacteristics similar to Ultrabook and high-end physical PCs

•Excellentuser-experiencecharacteristicseven with diverse use cases and burst-heavy workloads

Simple Architecture

•Desktopsandassociatedmanagementservers deployed on a single shared storage platform

•Efficientuseofstoragecapacitythroughbuilt-in SAN overprovisioning and inline deduplication

Extremely Fast Desktop Operations

•Userdesktoppoolsprovisionedquickly

•Enterpriseusecasesupandrunningthesamedayasrequested

•Newdesktops,applications,andhotfixesrolled out within a single maintenance window



OverviewAll VMware Horizon View desktops were hosted on the VCE Vblock Specialized System for Extreme Applications, which incorporates Cisco UCS server blades and EMC XtremIO flash-based arrays to ensure the performance and responsiveness needed to support massively scalable solutions.

VCE Vblock Specialized System for Extreme Applications

ThisVblockSystemisspecificallydesignedforsolutionssuchasvirtualdesktopinfrastructure(VDI).Currentlytargetedforgeneralavailabilityinthefourthquarterof2013,itenablesalowerpriceperdesktop,richuserexperience,flexibleVDIdeployment,andsimplifiedadministration.

It consists of a dedicated 6-chassis, 48-blade UCS blade environment. Infrastructure and management servers are isolated on four Cisco UCS C220 rack-mount servers, with storage for all virtual desktops and infrastructure servershostedonthesameEMCX-Brickstorage(twoX-Bricks)environment.UserdataandHorizonViewpersona data are directed to Isilon S-series NAS.

Adding an EMC Isilon for user data provides a simple, low-cost, file-based solution with good scale-out design. Putting user data in Isilon NAS and virtual desktops on XtremIO SAN increases desktop footprint capacity and value per desktop, and it also improves user experience by separating storage I/O workloads and directing them to different storage platforms. The configuration has the ability to grow in a linear fashion along with the environment.

Note: Testing was performed on a pre-release version of the Vblock Specialized System for Extreme Applications. Contact VCE directly for availability.

Infrastructure4 x Cisco UCS C220 M3

- Domain Controllers x 2- DHCP x 2- MS SQL x 2- vCenter x 2- Composer x 1- vC Ops for Horizon View x 1- vCenter Log Insight x 1- Horizon View x 4

8 x Isilon S-Series- User Data- Persona Data

2 x EMC XtremIO X-Bricks- Desktop- Infrastructure Virtual Machines

Desktop Hosts- 6 Cisco UCS Blade Chassis- 48 Cisco B200 M3 Blades (16 core, 256GB of RAM)Desktops

Figure 2: Infrastructure Overview



VMware Horizon View

VMware Horizon View provides unified access to virtual desktops and applications that run in a secure, centralized datacenter and are accessible from a wide variety of devices. Horizon View allows IT to manage desktops, applications, and data centrally while increasing flexibility and customization at the endpoint for the user. It enables levels of availability and agility of desktop services unmatched by traditional PCs at about half thetotalcostofownership(TCO)perdesktop.

Unlike other desktop virtualization products, Horizon View is a tightly integrated, end-to-end solution built on the industry-leading virtualization platform, VMware vSphere®. Horizon View allows customers to extend business continuity and disaster recovery features to their desktops and to standardize on a common platform, from the desktop through the datacenter to the cloud.

Storage Components

The VCE Vblock Specialized System for Extreme Applications uses EMC XtremIO storage arrays as primary storage for Horizon View virtual desktops and EMC Isilon NAS to store user data and Horizon View persona data.

EMC XtremIO XtremIOisahighlyscalable,all-flashstoragearraythatcombinesmulti-levelcell(MLC)flashwithsophisticatedwear leveling, data reduction, and write abatement technology to deliver extended flash endurance that makes the system both enterprise-reliable and reasonably priced.

The smallest unit of an XtremIO system is an X-Brick, and additional uniformly sized X-Bricks can be added forscalabilityinanXtremIOdesign.Asimpleuserinterfacereducesthetimerequiredtodesignandprovisionstorage. Enterprise resiliency is delivered through Fibre Channel connectivity, flash-specific dual-parity data protection, and storage presentation over the iSCSI protocol.

EMC Isilon NASThis reference architecture leverages an EMC Isilon as an optional add-on scale-out NAS component to the Vblock System. Isilon delivers increased performance for file-based data applications and workflows from a single file system. It can reduce, or even eliminate, the need to overprovision storage capacity or performance. As an additional component, EMC Isilon is sold separately.

Compute and Networking Components

The VCE Vblock Specialized System for Extreme Applications uses Cisco UCS blade enclosures, interconnects, and blade servers.

The Cisco UCS datacenter platform combines x86-architecture blade and rack servers with networking and storage access into a single system. Innovations in the platform include a standards-based, unified networkfabric,CiscoVirtualInterfaceCard(VIC),andCiscoUCSExtendedMemoryTechnology.Awire-once architecture with a self-aware, self-integrating, intelligent infrastructure eliminates the need for manual assembly of components into systems.

Cisco UCS B-Series two-socket blade servers deliver record-setting performance to a wide range of workloads. Based on Intel Xeon processor E7 and E5 product families, these servers are designed for virtualized applications and reduce CapEx and OpEx through converged network fabrics and integrated systems management.

Workload Generation and Measurement

LoginVSIisanindustry-standardtooldesignedtomeasurethemaximumcapacityofVDIinfrastructuresbysimulatinguniqueuserworkloads.Thesimulatedusersworkwiththesameapplicationsastypicalemployees,such as Microsoft Word, Excel, Outlook, and Internet Explorer. The results of several testing measurements are compiledinametricknownasVSImax,whichquantifiesthemaximumcapacityofVDIworkloadsrunningonagiven infrastructure while delivering an acceptable user experience.



Test ResultsTest results are summarized in the following sections. For further details, see Appendixes A through C.

Login VSI

We used the Login VSI workload-generation and measurement tool to generate and measure rigorous, realistic desktop workloads. Of the several types of Login VSI tests, two produced the most revealing findings.

VSImax and Horizon View Session ConcurrencyInTest1(7,000desktops,100percentconcurrency),examinationofbothsetsoftestresultsshowsaVSImaxof6,432.Atthislevelofsessionconcurrency,hostCPUresourcesarepushedtothepointofCPUsaturation,anduser experience starts to break down. This is not a viable, sustained level of CPU utilization because little CPU headroom is available for burst capability or failover. The sustained upper CPU utilization threshold for most production implementations is 85 percent. Sustained CPU utilization above 85 percent ordinarily causes a high CPU usage alarm in VMware vSphere.

Test2(7,000desktops,80percentconcurrency)showedmuchbetterhostCPUresourceutilization.All5,600sessions demonstrated excellent performance, and host CPU resources remained under 85 percent. All 7,000 desktops were powered on and available while the Login VSI workload ran on 80 percent of the available desktops.TypicalproductionVDIenvironmentsexhibitconcurrentdesktopusageof80percentoftotalavailable capacity.

Login VSI Test 1: 7,000-Seat Test with 100 Percent Session ConcurrencyHighlightsofTest1include:

•DesktopaccessusingthePCoIPprotocolandMediumWorkload(withFlashenabledbydefault)

•LoginVSIVSImaxof6,432

•Mixedhostperformance(CPUsaturatedbutmemoryusageunder50percent)

•Excellentdesktopperformance(allvDisklatencyunder0.5msforreadsandwrites)

•Peakof65,000IOPSonXtremIOStorage(75percentwrites,25percentreads)

Figure 3: Login VSI Score – Test 1



vCPU Usage

AveragePeak Average

Peak

AveragePeak Average

Peak

AveragePeak Average

Peak

AveragePeak Average

Peak

320 Mhz

570 Mhz

1050 Mhz

Mhz

vDisk Latency

.4 ms

.7 ms

1.1 ms

ms

vRAM Active

520 MB

690 MB

825 MB

MB

vNIC Usage XMIT/RCV

120/35 KBps

235/60 KBps

486/153 KBps

KBps

Figure 4: Desktop Performance – Test 1

Average Host Resource Utilization

Front End NetworkThroughput Capacity

Front End StorageThroughput Capacity

Memory Consumed

CPU Usage

0% 100%50%

Figure 5: Host Performance – Test 1

IOPS

IOPS

IOPS ms

5000 70005500 6000 6500 7000 7000 7000 7000380000.95

1

1.05

1.1

1.15

1.2

1.25

1.3

1.35

40000

42000

44000

46000

48000

50000

52000

X-Brick Performance

Latency (ms)

Figure 6: Storage Performance – Test 1



Login VSI Test 2: 7,000-Seat Test with 80 Percent Session ConcurrencyHighlightsofTest2include:

•5,600activesessions,withremainingdesktopspoweredonbutnotinuse

•TestingusingthePCoIPprotocolandMediumWorkload(withFlashenabledbydefault)

•LoginVSIVSImaxnotreached

•Excellenthostperformance(CPUandmemoryusageunder85percent,nocontention)

•Excellentdesktopperformance(allvDisklatencyunder0.5msforreadsandwrites)

•Peakof50,000IOPSonXtremIOstorage(75percentwrites,25percentreads)

Figure 7: Login VSI Score – Test 2

vCPU Usage

335 Mhz

545 Mhz

1020 Mhz

Mhz

vDisk Latency

.4 ms

.8 ms

1.0 ms

ms

vRAM Active

515 MB

695 MB

815 MB

MB

vNIC Usage XMIT/RCV

AveragePeak Average

Peak

AveragePeak Average

Peak

AveragePeak Average

Peak

AveragePeak Average

Peak

125/40 KBps

240/75 KBps

470/155 KBps

KBps

Figure 8: Desktop Performance – Test 2



Average Host Resource Utilization

Front End NetworkThroughput Capacity

Front End StorageThroughput Capacity

Memory Consumed

CPU Usage

0% 100%50%

Figure 9: Host Performance – Test 2

IOPS

IOPS

IOPS ms

3,500 5,6004,000 4,500 5,000 5,500 5,600 5,600 5,6000

0.2

0.4

0.6

0.8

1

1.2

1.4

40,000

41,000

42,000

43,000

44,000

45,000

46,000

47,000

48,000X-Brick Performance

Latency (ms)

Figure 10: Storage Performance – Test 2

Timing Tests

MeasuredintheHorizonViewManager™console,1,000desktopspoweredonin15minutes,and7,000desktops powered on in just over two hours.

Minutes

Desktops

500 1,500 2,500 3,500 4,500 5,500 6,500 7,000

20

40

60

80

100

120

140

0

Number of Desktops Ready in Horizon View

Figure 11: Desktop Startup Testing

R E F E R E N C E A R C H I T E C T U R E / 1 0


Linked-Clone Desktop Pool Deployment We observed fast consistent, and reliable desktop pool deployments even with background workload on the compute and shared storage environment. With 5,000 desktops deployed and powered on, measurement of a 1,000-desktoplinked-clonepooldeploymentproducedthefollowingresults:

Minutes

Desktops

100 200 300 400 500 600 700 800 900 1,000

10

20

30

40

50

60

70

80

0

1,000-Seat Linked Clone Pool Deployment

Figure 12: Pool Deployment Measurements

Linked-Clone Desktop Pool Recompose We observed similarly fast, consistent, and reliable desktop pool recompose operations. With 5,000 desktops deployedandpoweredon,wemeasureda1,000-desktoplinked-clonepoolrecomposeandobservedthefollowingresults:

Minutes

Desktops

100 200 300 400 500 600 700 800 900 1,000

20

40

60

80

100

120

0

1,000-Seat Linked Clone Pool Recompose

Figure 13: Pool Recompose Measurements



Storage Capacity

The volume capacity configured and presented to vSphere totaled 59.9TB. The actual storage footprint for 7,000desktopsandinfrastructureserverswasapproximately6.24TB.DuetotheefficiencyoftheXtremIOinlinedata reduction capability, desktop and server virtual machines occupied a physical storage footprint of only 3.5TB.Thisrepresents:

•Overallefficiencyratio:24:1

•Deduplicationratio:2.6:1

•vSphereThinProvisioningsavings:89%

Overall E�ciency 24:1- Deduplication Ratio 2.6:1- Thin Provisioning Savings 89%

Volume Capacity 59.9TB

Physical Capacity 14.9TB

Storage

3.51TB Physical Space Used

53.7TB Free6.24TB Used

11.4TB Free3.51TB Used

Figure 14: EMC XtremIO X-Brick Capacity Display

This massive savings in storage capacity is a clear representation of how the EMC XtremIO storage platform can drivedownthestoragefootprintandassociatedcostsinamedium-tolarge-scaleVDIdeployment.



System ConfigurationsThe following sections describe in detail how the components of this reference architecture were configured.

vSphere Cluster Configurations

Fordesktopvirtualmachines,wedeployedtwovSphereclusterswith24nodes(withN+2HAreadiness).Formanagementandinfrastructurevirtualmachines,wedeployedasinglefour-nodecluster(withN+1HAreadiness).

Cluster 1 – 3,500 Desktops

Cluster 0 – Management / Infrastructure Servers

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host


Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Figure 15: Cluster Configurations

Networking Configurations

Eachfabricinterconnecthasmultipleportsreservedfor10GbEports.TheseconnectionsareformedintoaportchanneltotheCiscoNexusswitchandcarryIPtrafficdestinedforthedesktopnetwork10GbElinks.Inaunifiedstorage configuration, this port channel can also carry IP traffic to the X-Blades within the storage layer.

TheCiscoNexus5548UPswitchesinthenetworklayerprovide10GbEIPconnectivitybetweentheinfrastructure and the outside world. In unified storage architecture, the switches connect the fabric interconnects in the compute layer to the XtremIO X-Bricks in the storage layer.



In segregated architecture, the Cisco Nexus 5548UP series switches in the network layer provide Fibre Channel (FC)linksbetweentheCiscofabricinterconnectsandtheEMCXtremIOstoragearray.TheseFCconnectionsprovide block-level devices to blades in the compute layer.

Cisco Nexus 5548 Cisco Nexus 5548

EMC XtremIOX-Brick

UCS Blade Server

Fabric Interconnect Fabric Interconnect

Figure 16: Cisco Networking Infrastructure

IP Network ComponentsTosupporttheEthernetandSANrequirementsinthetraditionalsegregatednetworkarchitecture,twoCiscoNexus5548UPswitchesprovideboth10GbEEthernetandFibreChannel(FC)connectivity.

TheCiscoNexus5548UPswitcheshave32integrated,low-latencyunifiedports,eachprovidingline-rate10GBEthernet or FC connectivity. The Cisco Nexus 5548UP switches each have one expansion slot that can be populatedwitha16-portunifiedportexpansionmodule.

Storage Configurations

We used EMC XtremIO X-Bricks for linked-clone system disk storage, and we redirected user data and user persona to an EMC Isilon NAS storage appliance to validate this reference architecture. This approach mirrors a real-world implementation. Separating the storage workloads between system disks and user data and persona allowsaVDIarchitecttomatchtheuniquecharacteristicsoftheseparateworkloadstothestoragearraybestsuited to accommodate them.

We used two X-Bricks for this testing. The two systems were connected by InfiniBand switches and yielded 14TBrawcapacityandapproximately70TBusablecapacity,assuming5:1deduplicationefficiency.

ThefollowingcomponentsmadeupthedualX-Brickdeployment:

•TwoX-Brickdiskarrayenclosures

•FourX-Brickcontrollers

•Twobatterybackupunits

•TwoInfiniBandswitches



2 Battery Backup Units

X-Brick Disk Array Enclosure

X-Brick Disk Array Enclosure

2 X-Brick Controllers

2 X-Brick Controllers

2 In�niBand Switches

Figure 17: X-Brick Deployment

X-Brick Storage Configurations in vSphereModestsettingchangesarerequiredinvSpheretosupportEMCX-Brickstoragesystems:

•Adjustthenumberofoutstandingstoragetargetcommandsto256.

•Changethenumberofconsecutivestoragecommandsto64.

•ChangetheHBAqueuedepthsettingsto256foreachhost’sHBAs.

•ChangethenativestoragemultipathpolicyinvSpheretoRoundRobin.

Note: SeeAppendixB(TestMethodology)fordetailsonchangingthesesettings.

Each24-nodeclusterwaszonedandmaskedtoallowaccessto242TBdatastoresandasingle500GBdatastore for pool masters.


Clones 2TB

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Host Host Host Host

Clones 2TB Clones 2TB Clones 2TB

Clones 2TB Clones 2TB Clones 2TB Clones 2TB

Clones 2TB Clones 2TB Clones 2TB Clones 2TB

Clones 2TB Clones 2TB Masters 500GB

Figure 18: 24-Node Cluster Detail

The four-node management and infrastructure cluster was zoned and masked to allow access to a single 2TB datastore.



vSphere Configurations

We deployed separate virtual datacenters, one for management and monitoring servers and one for hosted desktops. This is the standard, recommended configuration for most production Horizon View deployments.

OS

APPvCenter for

Infrastructure

Server Virtual Machines Desktop Virtual Machines

vCenter for Desktops

OS

APP

vCenter - VDI

OS

APP

v Center SSO

OS

APP

v Center Inventory

OS

APP

Log Insight Server

OS

APP

View Composer

OS

APPOS

APP

DCs

OS

APPOS

APP

DHCP

OS

APPOS

APP

SQL Servers

OS

APPOS

APP

vC Ops Servers

OS

APPOS

APP OS

APPOS

APP

Connection Server Cluster

OS

APP

Pool 1

1,000 Desktops

Pool 2

1,000 Desktops

Pool 3

1,000 Desktops

Pool 4

1,000 Desktops

Pool 5

1,000 Desktops

Pool 6

1,000 Desktops

Pool 7

500 Desktops

Pool 8

500 Desktops

TemplateMasters

Figure 19: Virtual Datacenters

Infrastructure and Management Servers

We deployed two VMware vCenter Server™ instances, one for desktops and one for infrastructure and management servers, in conformance with Horizon View 5.2 architecture planning guidelines.

Note: AsofHorizonViewversion5.2,VMwaresupportsmanagingupto10,000desktopswithasinglevCenter5.2 server instance. This is a significant revision from prior versions.

AlladditionalvCenter™roles(inventory,SSO,vCenter)forthedesktopvCenterweredivestedtoseparateservers to avoid any resource contention that might have resulted from combining roles on a busy vCenter Server.

All server resources were sized according to the current best practices from VMware. They are listed in the followingtable:

SERVER ROLE VCPU RAM (GB) STORAGE (GB) OS

DomainController1 2 4 40 WindowsServer200864-bitR2

DomainController2 2 4 40 WindowsServer200864-bitR2

SQLServer1 2 6 140 WindowsServer200864-bitR2

http://www.vmware.com/files/pdf/view/vmware-horizon-view-best-practices-performance-study.pdf



SERVER ROLE VCPU RAM (GB) STORAGE (GB) OS

SQL Server 2 2 6 140 WindowsServer200864-bitR2

Horizon View Composer™ Server

4 6 40 WindowsServer200864-bitR2

vCenter Server – Infrastructure


vCenter Server – Desktops


Horizon View Connection Server1


Horizon View Connection Server 2


Horizon View Connection Server3


Horizon View Connection Server 4


VMware vCenter Operations Manager™ for Horizon View Analytics Server

16 20 275 SLES11–64-bit

VMware vCenter Operations Manager for Horizon View UI Server

8 12 145 SLES11–64-bit

vCenter Log Insight 2 8 140 SLES11–64-bit

DHCPServer1 1 4 40 WindowsServer200864-bitR2

DHCPServer2 1 4 40 WindowsServer200864-bitR2

vCenter Inventory Service –Desktops


vCenterSSO–Desktops 4 8 55 WindowsServer200864-bitR2

Table 1: Server Resource Sizing



Horizon View Configuration

TheHorizonViewinstallationincludedthefollowingcoresystems:

•Fourconnectionservers

•DedicatedvCenterwithrolessplittoseparateserversasfollows:

- vCenter

-vCentersinglesign-on(SSO)

- vCenter Inventory Service

•ViewComposerrunningonaseparateserverfromvCenter

OS

APP

vCenter

OS

APP

vCenter SSO

OS

APP

vCenter Inventory

OS

APP

Log Insight Server

OS

APP

View Composer

OS

APPOS

APP

SQL Servers

OS

APPOS

APP

vC Ops Servers

OS

APPOS

APP OS

APPOS

APP

Connection Server Cluster

Figure 20: Horizon View Installation

Note: Security servers were not used during this testing.

Basic Horizon View SettingsThe Horizon View global settings are summarized in Table 2.

ATTRIBUTE SPECIFICATION

GLOBAL POLICIES

Multimediaredirection(MMR) Allow

USB access Allow

Remotemode Allow

PCoIP hardware acceleration Allow – medium priority

GLOBAL SETTINGS

Session timeout 600(10hours)

SSO Always Enabled




View Administrator session timeout 120minutes

Automatic status updates Enabled

Pre-login message None

Warning before forced logoff None

GLOBAL SECURITY SETTINGS

Re-authenticatesecureconnectionsafternetworkdisrupt Enabled

Message security mode Enabled

Enable IPSec for Security Server pairing Enabled

DisableSSOforLocalMode Disabled

Table 2: Horizon View Settings

Test Pool ConfigurationWedeployedatotalofeightlinked-clonepools,asillustratedinFigure21.

Pool 1

1,000 Desktops

Pool 2

1,000 Desktops

Pool 3

1,000 Desktops

Pool 4

500 Desktops

Pool 5

1,000 Desktops

Pool 6

1,000 Desktops

Pool 7

1,000 Desktops

Pool 8

500 Desktops

Figure 21: Test Pool Deployment



Desktop Pool ConfigurationsLinked-clone pool configurations conform to a typical knowledge worker profile, with the exception of the storageovercommitpolicy.TheconfigurationsaredetailedinTable3.


Pool type Linked Clone

Persistence Non Persistent

PoolID DesktopPool###

Displayname DesktopPool###

FolderID /

Separate datastores for replica and OS? Not selected

State Enabled

Connection Server restrictions None

Remotedesktoppowerpolicy Take no power action

Auto logoff after disconnect Never

User reset allowed False

Multiple sessions per user allowed False

Deleteorrefreshdesktoponlogoff Never

Displayprotocol PCoIP

Allow protocol override False

Max number of monitors 2

Max resolution 1920x1200

HTML Access Not selected

Flashqualitylevel Medium

Flash throttling level Moderate

Enable provisioning Yes

Stop provisioning on error No




Provisioning timing All up front

Use View Storage Accelerator No

Reclaimvirtualmachinediskspace Yes,2GB

Storage overcommit policy Aggressive*

* Storage overcommit policy must be set to Aggressive with any storage platform that has an inline deduplication or data-reduction capability.

Table 3: Desktop Pool Configurations

Cluster to Datastore AssignmentsEach1,000-seatpoolwasdeployedacrossfourdatastoresassummarizedinTable4.

POOL NAME DATASTORE NAME DATASTORE SIZE

Cluster1–Pool1 Datastore1 2TB

1,000desktops Datastore2 2TB

Datastore3 2TB

Datastore4 2TB



Datastore7 2TB

Datastore8 2TB



Datastore11 2TB

Datastore12 2TB


500 desktops Datastore14 2TB



POOL NAME DATASTORE NAME DATASTORE SIZE



Datastore17 2TB

Datastore18 2TB

Cluster2 – Pool2 Datastore19 2TB


Datastore21 2TB

Datastore22 2TB



Datastore25 2TB

Datastore26 2TB

Cluster2 – Pool4 Datastore27 2TB

500 desktops Datastore28 2TB

Table 4: Cluster to Datastore Assignments



Test Image ConfigurationWe configured the virtual hardware of the master desktop virtual machine according to standard Login VSI specifications. It is important to note that in production deployments, virtual machine configurations vary basedonindividualuse-caserequirements.


DesktopOS Microsoft Windows 7, 64-bit

VMware virtual hardware Version 9

VMware Tools version 9.221(uptodate)

Virtual CPU 1

Virtual memory 2048MB

OS pagefile 1.5GBstartingandmaximum

vNICs 1

Virtualnetworkadapter1 VMXNet3Adapter

Virtual SCSI controller 0 LSI Logic SAS

VirtualDisk–VMDK 40GB

VirtualFloppyDrive1 Removed

VirtualCD/DVDDrive1 Removed

VMware Horizon View Agent VMwareHorizonViewAgent5.2build987719

Installed applications (per Login VSI standards)

MicrosoftOffice2010

AdobeAcrobatReader10

AdobeFlashPlayer10

AdobeShockwavePlayer10

BullzipPDFPrinter

FreeMind

Kid-Key-Lock

Table 5: Test Image Configuration

ThemasterimageusedinthistestenvironmentunderwentVDIoptimizationasdefinedintheVMware Horizon View Optimization Guide for Windows 7. VMware strongly recommends that the Windows image be optimized when master images are prepared for use with VMware Horizon View.

http://www.loginvsi.com/documentation/v3/installation/target-setup

https://www.vmware.com/files/pdf/VMware-View-OptimizationGuideWindows7-EN.pdf

https://www.vmware.com/files/pdf/VMware-View-OptimizationGuideWindows7-EN.pdf



EMC XtremIO Storage Configurations

As we would with any shared storage array, we followed VMware and storage vendor best practices for integration with vSphere. Execute the following VMware ESXi™ parameter adjustments to properly configure ESXitoworkwithEMCXtremIOX-Bricks:

1. Connect to the VMware ESX® host via the vSphere Client (directly or through the vCenter Server).

2. Click the Host icon in the left pane and click Advanced Settings.

3. Click the Disk section.

4. Find the Disk.SchedNumReqOutstanding parameter and change it to the Max value (256).

5. Find the Disk.SchedQuantum parameter and change it to the Max value (64).

6. Click OK to apply the changes.

Figure 22: ESXi Parameter Adjustments




vStorage API for Array Integration (VAAI) SettingsVAAI is a vSphere API that lowers desktop provisioning time from hours down to minutes by performing vSphere operations such as virtual machine provisioning and cloning within an array that supports VAAI. The XtremIO storage system fully supports VAAI. VAAI is enabled by default in vSphere version 5.x. No further action is requiredforVAAItobeusedwithXtremIOstorage.

HBA Queue Depth AdjustmentToadjustthequeuedepth,usethefollowingsteps:

1. Connect to ESX host shell as root.

2. Verify which HBA module is currently loaded by entering one of the following commands:

• For Qlogic: esxcli system module list | grep qla

• For Emulex: esxcli system module list | grep lpfc

For example, from a host with a Qlogic HBA:

# esxcli system module list | grep qla qla2xxx true true

3. To adjust the HBA queue depth, run one of these two commands:

• For Qlogic: esxcli system module parameters set -p ql2xmaxqdepth=256 -m qla2xxx

• For Emulex: esxcli system module parameters set -p lpfc0_lun_queue_depth=256 -m lpfc820

4. Reboot the ESX host.

5. Connect to the ESX host shell as root.

6. Run the following command to confirm the queue depth adjustment:

# esxcli system module parameters list -m <driver>

For example, from a host with a Qlogic HBA with queue depth set to 256:

# esxcli system module parameters list -m qla2xxx | grep ql2maxqdepth ql2maxqdepth int 256 Maximum queue depth to report for target devices

ForfurtherinformationaboutadjustingHBAqueuedepthwithESX,refertoVMware Knowledge Base article 1267.

http://kb.vmware.com/selfservice/microsites/search.do?cmd=displayKC&docType=kc&externalId=1267&sliceId=2&docTypeID=DT_KB_1_1&dialogID=58468502&stateId=1%200%2058486714



Native vSphere Storage MultipathingXtremIO supports the native multipathing technology that is part of the VMware vSphere suite.

For the best performance, set the native Round Robin path selection policy on XtremIO volumes presented to ESXi to ensure optimal distribution and availability of load among I/O paths to the XtremIO storage.

Figure23illustrateshowtomodifythepathselectionpolicyfromthedefaultFixed path selection policy to the Round Robin policy.

Figure 23: Path Selection Policy

ConclusionWhen designing and estimating capacity for this VMware Horizon View large-scale reference architecture, weadheredcloselytopublishedVMwaresizingandestimationstandards.Designanddeploymentwerestraightforward, time-efficient, and trouble-free, yielding rapid pool deployment and recompose times that showed desktop maintenance was feasible, even during the middle of a production workday, and even with extensive background workloads running simultaneously. The combination of simplicity and performance is without peer in the industry, and it surprised even those of us on the reference architecture team—in spite of having expected these results up front!

OurtestresultsdemonstratethatitispossibletodeliveranUltrabook-qualityuserexperienceatscaleforeverydesktop,withheadroomforanydesktoptobursttothousandsofIOPSasrequiredtodriveuserproductivity,thanks to the EMC XtremIO storage platform, which provides considerably higher levels of application performance and lower virtual desktop costs than alternative platforms. The high performance and simplicity of the EMC XtremIO array and the value-added systems integration work provided by VCE as part of the Vblock design contributed significantly to the overall success of the project.



About the AuthorsTristanToddisanEUCArchitectintheVMwareEnd-UserComputingTechnicalEnablementGroup.Hehasextensive customer, field, and lab experience with VMware End-User Computing and ecosystem products.

ShreeDasisaSolutionsArchitectatVCEresponsiblefordesignanddevelopmentofcloudandvirtualizationreferencearchitecturesonVblockplatforms.Hisareasofexpertiseincludedatacenterdesign,VDI,multi-tenancy,VMwarevCloudDirector,andIaaS.

ShridharDeuskarisaSolutionsArchitectwithEMCXtremIO.Hehasover15yearsofexperienceintheindustry,including virtualization and storage technologies.

DennisGeerlingsisSupportManagerandLeadConsultantatLoginVSI.Intheseroles,hebringstocustomersandpartnersyearsofexperienceinenterprisedesktopandVDIdesignanddeployment.Inaddition,heisanactive developer on the Login VSI engineering team.

Acknowledgments

This reference architecture is the result of collaboration between VMware, EMC, VCE, and Login VSI. VMware recognizesitspartners’generosityinprovidingequipment,time,andexpertise,withoutwhichthisprojectwould not have been possible.

EMC is a global leader in enabling businesses and service providers to transform their operations and deliver informationtechnologyasaservice(ITaaS).Fundamentaltothistransformationiscloudcomputing.Throughinnovative products and services, EMC accelerates the journey to cloud computing, helping IT departments to store, manage, protect and analyze their most valuable asset, information, in a more agile, trusted, and cost-efficient way. For more information, visit http://www.emc.com.

VCE, formed by Cisco and EMC with investments from VMware and Intel, accelerates the adoption of converged infrastructure and cloud-based computing models that dramatically reduce the cost of IT while improving time to market for its customers. Through the Vblock Specialized System for Extreme Applications, VCE delivers theindustry’sonlyfullyintegratedandfullyvirtualizedcloudinfrastructuresystem.Formoreinformation,visithttp://www.vce.com.

Login VSI is an international software company focused on helping both end users and vendors of virtual desktop infrastructures to design, build, implement, and protect the best performing hosted desktop infrastructures possible. For more information, visit http://www.loginvsi.com.

http://www.emc.com

http://www.vce.com

http://www.loginvsi.com



ReferencesVMware

VMware Horizon View Architecture Planning

VMware Horizon View Installation

VMware Horizon View Administration

VMware Horizon View Security

VMware Horizon View Integration

Storage Considerations for Horizon View 5.2

Security Considerations for Horizon View 5.2

Server and Storage Sizing Guide for Windows 7

Windows 7 and Windows 8 Optimization Guide for Horizon View Virtual Desktops

Horizon View 5 with PCoIP Network Optimization Guide

Horizon View 5.2 Performance Data and Best Practices

PowerShell Integration with Horizon View

Workload Considerations for Virtual Desktop Reference Architectures

EMC

XtremIO Overview

XtremIO VDI Solutions Brief

XtremIO Customer Case Study

Cisco

UCS Solution Brief

UCS and Horizon View Solutions Portal

Cisco UCS SmartPlay Configurations for VMware Desktop Workloads

VCE

VCE Vblock Specialized System for Extreme Applications Product Overview

VCE End User Computing Solutions Portal

Login VSI

Login VSI Technical Documentation

Login VSI Workload Overview

Login VSI Deployment Guide

Login VSI Operations Guide

Cisco Login VSI Performance Study for Horizon View

http://www.vmware.com/support/pubs/view_pubs.html





http://www.vmware.com/files/pdf/view_storage_considerations.pdf

http://www.vmware.com/files/pdf/techpaper/vmware-horizon-view-security.pdf

http://www.vmware.com/files/pdf/view/Server-Storage-Sizing-Guide-Windows-7-TN.pdf

http://www.vmware.com/files/pdf/VMware-View-OptimizationGuideWindows7-EN.pdf

http://www.vmware.com/files/pdf/view/VMware-View-5-PCoIP-Network-Optimization-Guide.pdf


http://www.vmware.com/files/pdf/vmw-powershell-integration-view5.pdf

http://www.vmware.com/files/pdf/VMware-WP-WorkloadConsiderations-WP-EN.pdf

http://www.us.emc.com/collateral/brochure/emc-xtremio-brochure.pdf

http://www.us.emc.com/collateral/solution-overview/emc-xtremio-vdi-so.pdf

http://www.us.emc.com/collateral/analyst-reports/xtremio-green-enterprise-it-case-study.pdf

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns944/le_32302_sb_itagility.pdf

http://www.cisco.com/en/US/netsol/ns1136/index.html

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns836/ns978/solution_overview_c22-726854_ns1136_Networking_Solution_Solution_Overview.html

http://www.vce.com/products/specialized/extreme-apps

http://www.vce.com/solution/applications/end-user

http://www.loginvsi.com/documentation/v3

http://www.loginvsi.com/documentation/v3/performing-tests/workloads

http://www.loginvsi.com/documentation/v3/installation

http://www.loginvsi.com/documentation/v3/performing-tests

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns836/ns978/white_paper_c11-727392.pdf



Appendix A (Login VSI Detailed Scoring Methodology)TheLoginVSIMediumWorkload(withFlash)isacommonindustry-standardbenchmarkforVDIandSBCtesting.Thetestinvolvesthefollowingconsiderations:

•ThisworkloadsimulatesamediumknowledgeworkloadusingOffice,InternetExplorer,andPDF.

•Onceasessionhasbeenstarted,themediumworkloadrepeatsevery12minutes.

•Duringeachlooptheresponsetimeismeasuredeverytwominutes.

•Themediumworkloadopensuptofiveapplicationssimultaneously.

•Thetyperateis160msforeachcharacter.

•Approximatelytwominutesofidletimeisincludedtosimulatereal-worldusers.

EachLoginVSIloopinvolvesthefollowingapplicationoperations:

•Outlook2007/2010–Browse10messages.

•InternetExplorer–Oneinstanceisleftopen(BBC.co.uk),oneinstanceisbrowsedtoWired.comandLonelyplanet.com.

•Word2007/2010–Oneinstancetomeasureresponsetime;oneinstancetoreviewandeditdocument.

•BullzipPDFPrinterandAdobeAcrobatReader–TheWorddocumentisprintedandreviewedtoPDF.

•Excel2007/2010–Averylargerandomizedsheetisopened.

•PowerPoint2007/2010–Apresentationisreviewedandedited.

•7Zip–Usingthecommandlineversion,theoutputofthesessioniszipped.



Appendix B (Test Methodology)This reference architecture used both deployment and recompose timing and Login VSI testing.

Deployment and Recompose Timing Tests

DeploymentandrecomposetimemeasurementsareoftenoverlookedinHorizonViewreferencedesignsbecausetheyareconsideredinfrequentevents.Deployandrecomposetimesarecriticalbeyondtheinitialsetupforseveralreasons:

•DeploymentandrecomposeeventshaveadirectimpactontheITteam’sagilityandresponsiveness.Considertheexampleofalargecompanythatcompletesanacquisitionandmustrolloutacorporatedesktopandapplicationstoallthenewlyacquiredemployees.Deploymenttimesaredirectlycorrelatedtothespeedatwhich this can be accomplished.

•Linked-cloneVDIimagesarerecomposedandredeployedperiodicallytoapplyOSpatchesorapplicationupdates,whichrequiresamaintenancewindow.Therecomposeoperationmustbecompleted,therefore,asrapidly as possible.

•Manycustomersemployaregular,automatedrecomposepolicybysettingdesktopstoRefresh on Logoff.

Note: All pool deployment and recompose timing tests in this reference architecture included not only the duration of the master image cloning process but virtual desktop customization, power on, domain registration, and View Agent registration. This is a more realistic test than simple creation of the image clones.

Login VSI Testing

LoginVirtualSessionIndexer(LoginVSI)istheindustry-standardbenchmarkingtoolformeasuringtheperformanceandscalabilityofcentralizeddesktopenvironmentssuchasVirtualDesktopInfrastructure(VDI)andServer-BasedComputing(SBC).

We used Login VSI to generate a reproducible, real-world test case that simulated the execution of various applications, including Microsoft Internet Explorer, Adobe Flash video, and Microsoft Office applications, to determine how many virtual desktop users, or sessions, each solution could support.

Login VSI was configured to run a medium workload against a Horizon View pool of virtual desktops, with the testssetuptologusersintovirtualdesktopsincrementallyevery30secondsbetweensessionperphysicalhost(blade).

Login VSI measured the total response time of all the applications from each session and calculated the VSI Index by taking the average response times and dropping the highest and lowest two percent.

Theaverageresponsetimeofthefirst15sessionsdeterminedabaseline.Asmoresessionsbegantoconsumesystem resources, response time degraded, and the VSI index increased until it went above the VSImax (VSImaxisbaselinex125%+3000ms).Whenthisconditionwasmet,thebenchmarkrecordedaLoginVSImax,which is the maximum number of sessions that the platform could support.

Onceloggedin,eachsessionremainedactiveforthedurationofthetest,andforatleast15minutes.Notreaching VSImax is an indication of satisfactory user response at the predetermined user count.

Duringtesting,LoginVSIsessionswereinitiatedbylaunchers(simulateduserendpoints)thatranonseparatecomputeandstorageinfrastructure.Atotalof140launcherswereutilized,eachrunning50sessions.EachlauncherwasconfiguredwithtwovCPUsand6GBofvRAM,followingLoginVSIsizingguidelines.

VMware, Inc. 3401 Hillview Avenue Palo Alto CA 94304 USA Tel 877-486-9273 Fax 650-427-5001 www.vmware.comCopyright © 2013 VMware, Inc. All rights reserved. This product is protected by U.S. and international copyright and intellectual property laws. VMware products are covered by one or more patents listed athttp://www.vmware.com/go/patents. VMware is a registered trademark or trademark of VMware, Inc. in the United States and/or other jurisdictions. All other marks and names mentioned herein may be trademarks of their respective companies. Item No: VMW-WP-HORZVWLSREFARCH-20131025-WEB


Appendix C (Bill of Materials for Test Configuration)The test configuration bill of materials is summarized in Table 6.

AREA COMPONENT QUANTITY

Host hardware UCS5108BladeChassis 6

CiscoUCSB200-M3(2xIntelE5-2680,256GBRAM) 48

CiscoC220-M3(2xIntelE5-2680,256GBRAM) 4

Storage hardware EMC X-Bricks 2

IsilonS200NASnode(7.2TBcapacity) 8

InfiniBand36-portswitch 2

Network hardware CiscoNexus3048Switch 2

Cisco Nexus 5548P Switch 2

Cisco Nexus 6248P Fabric Interconnect 2

Software inventory VMwareESXi5.1Update1,build1117900 4

VMware vCenter 2

VMware Horizon View 7,000 users

VMware vCenter Operations Manager for Horizon View 8,000 virtual machines

VMware vCenter Log Insight 2 vCenter instances

Microsoft Server 2008 15

Microsoft SQL Server 2008 2 Table 6: Bill of Materials

Documents

Vmware Horizon View Large Scale Reference Architecture